I don't know of any pre made tools, but if you go with Deepbit for a week or so they have a really simple JSON API which you can write tools for yourself. I wrote a simple R script to dl all the data every 2 minutes. Then after a few hours of data collection:
1. Plot blocks vs day (time is reported in Unix days)
2. Do a linear line of fit, y=a*c +b.
This gives you blocks per day and this is great for comparing various flags and settings.
If you want to convert to Mhps:
Mhash/sec = (blocks*1.193)/24
eg: for my 6990 running DM for 20 hours with no flags but overclocked, I get the following line of fit: y = 13201x - 2E+08
so block rate = 13201 blocks/day
Mhash/sec = 13201*1.193/24 = 656 Mhps
you can do the same with your invalids: y = 105*x-833564
invalids per day = 105
%invalid = 105/13201 = 0.79%
You should be able to put all that into the script you write to get the JSON data so you just get an output at the end of your run.
It's tempting to just divide total blocks in a given time period by time, but if you have a lucky hour at the end it throws off your result. Better to use linear modelling over the whole time period.
Good luck! Post your results when you get them - always good to see how flags and OC and UC affect Mhps.